Anthropic's Trillion-Dollar Window

Programming note: ARPU will be back next Friday with a look at the AI boom from Asia.

The Art of Raising Money

There is an old rule in corporate finance: you raise money when you can, not when you need it.

Anthropic is currently executing on that principle at historic scale. The maker of Claude is preparing for what could easily become the largest initial public offering in human history, aiming for a valuation somewhere between $1.5 trillion and $2 trillion.

Or at least it was trying to.

On Thursday, the IPO window Anthropic has been racing to get through began to visibly narrow. The Nasdaq fell 5.5% into correction territory, and the semiconductor index suffered its worst day since March 2020.

The sell-off was driven by fears that Tuesday's Democratic midterm sweep, together with a raft of ballot measures targeting data centers, could derail the AI spending boom that has powered tech stocks for three years. On prediction markets, the odds of Anthropic going public by the end of November fell from 84% to 60% in a single day.

Anthropic's IPO timing is fascinating for another reason, too. Even before Thursday, the public markets had been an absolute bloodbath for literally everyone else in the AI universe.

CoreWeave's stock had already shed 40% amid a credit panic. A string of IPO hopefuls, from Australian cloud provider Firmus to UK-based Nscale, had quietly canceled or postponed their stock market debuts. Even OpenAI had looked at the carnage and decided to postpone its own listing until at least 2027.

Yet Anthropic is charging through the front door.

This isn't because the company is running on fumes. Its S-1 prospectus discloses a comfortable $146 billion in total available liquidity, paired with some of the most breathtaking top-line growth Silicon Valley has ever printed: $33 billion in revenue over the first nine months of the year.

So why the hurry? Thursday showed how quickly the market can turn on AI. But the S-1 filing suggests Anthropic had reasons to move fast well before the market wobbled. Read past the top-line numbers and it reveals a business standing at an important crossroads. The extraordinary growth of early 2026 is beginning to normalize, pricing competition is setting in, and buried in the footnotes is an $895 billion collection of infrastructure commitments that will require tens of billions of dollars in future spending.

The Quarter That Changed the Story

To understand the attraction of an IPO now, consider the rather dramatic difference between Anthropic's second and third quarters.

On paper, the first half of the year was a triumph. For two years, skeptics had dismissed frontier AI labs as glorified money furnaces that set fire to billions in compute without ever finding a sustainable business model. In the second quarter, Anthropic delivered the ultimate counterpunch: it posted a GAAP operating profit of $974 million, while sequential quarterly revenue growth surged by an eye-watering 144%.

It looked like the start of an exponential, high-margin software empire.

What the filing doesn't say out loud is that the first half of 2026 was the absolute peak of the corporate "token-maxing" craze. This was the brief window where companies ordered their software developers to use AI for everything with virtually no budget constraints. At Meta, engineers competed on an internal leaderboard called Claudeonomics, handing out titles like "Token Legend" and "Cache Wizard" to staff who burned through a million dollars of Claude tokens a month, while management cheered them on.

That enthusiasm was certainly welcome news for companies selling AI tokens. But over the summer, corporate CFOs finally looked at their cloud invoices and remembered that companies are supposed to make money. Amazon sent out a memo politely warning engineers that token usage is not a measure of productivity, a sentence nobody had previously needed to write down. Corporate spending slowed; and enterprises started routing routine queries to cheaper open-weight models.

The hangover showed up in Anthropic's third-quarter disclosures. Revenue was still massive, hitting nearly $17 billion. But sequential growth plummeted from 144% to 47%. Now, growing 47% in three months would normally prompt a celebratory round of golf. But when you are pitching Wall Street on a valuation roughly twice the size of Switzerland's entire economy, a sudden deceleration from a sprint to a jog is the kind of thing that makes investment bankers sweat.

Profitability also came under scrutiny. As computing expenses climbed, Anthropic swung from that $974 million operating profit in Q2 back to a $567 million operating loss in Q3. And the filing confirmed that price cuts and contractual discounts shaved roughly $600 million off third-quarter revenue.

None of this suggests Anthropic is in trouble. Its customer base continues to expand rapidly, and its operating performance has improved enormously over the past year.

But it does suggest the company is entering a different phase of growth. The remarkable financial acceleration of early 2026 is giving way to a more complicated combination of enterprise procurement discipline, pricing competition and infrastructure costs.

That new phase now lands at an awkward moment. After Thursday, strategists are describing a handoff in AI stocks from growth investors, who are heading for the exits, to value investors, who want healthy cash flows and don't think these stocks are cheap yet. A company whose growth rate just fell by two-thirds in a single quarter, and which slipped back into an operating loss, is asking to be priced right in the middle of that handoff. If you're going to establish a trillion-dollar public valuation, there's an obvious attraction to doing so while the recent growth figures still look extraordinary, and before the investors who loved them have all left the room.

Selling the Future Before the Bill Arrives

The second reason the timing looks attractive lives in the prospectus's less glamorous discussion of contractual obligations.

Over the past ten months, Anthropic has signed a dizzying array of hardware commitments: $110 billion with Amazon, $121 billion with Google, $120 billion with SpaceX, and $161 billion with Broadcom (Broadcom, Anthropic's largest single counterparty, fell 10% on Thursday alone.) In total, the company is on the hook for $895 billion in future compute and infrastructure spending.

These are not immediate debts, and they stretch across multiple years. But many of these contracts involve "take-or-pay" provisions, meaning Anthropic has to pay for the capacity whether its customers show up to use it or not.

And the near-term bill is hardly trivial: nearly $70 billion falls due in the next twelve months alone.

This week's midterm elections complicate the picture further, adding a variable that looked like a distant risk factor when the S-1 went public.

If newly elected governors and a new Congress slow data-center permitting, the cloud capacity Anthropic has already contracted for could arrive late, even as some of its financial commitments remain firmly in place. But there is also a counter-scenario worth taking seriously: if new capacity becomes scarce, the compute Anthropic has already locked up could start to look less like a liability and more like a moat.

Investors now have to decide which of those two stories they are buying.

Capital Without the Coupon

Capital Without the Coupon

And this is where Anthropic's IPO becomes particularly interesting.

According to the a report from the Financial Times, Blackstone now expects Anthropic to receive a double-B credit rating following its listing—a "junk" rating that could make borrowing considerably more expensive. That's a marked reversal from Blackstone's earlier expectation that the company would achieve investment-grade status.

Which makes the equity market an especially attractive alternative.

At a hypothetical $2 trillion valuation, Anthropic could raise $100 billion by issuing just 5% of its equity. That's an extraordinary amount of permanent capital for a relatively modest slice of ownership, without the interest payments and maturity dates that have become such an uncomfortable feature of the AI infrastructure boom.

And this brings us back to the old rule in corporate finance: raise money when you can, not when you need it. Anthropic may still find plenty of willing investors. Whether they'll be quite as generous as they were a few weeks ago is another matter.

Signal Stack

The operating reality beneath the headlines.

• OpenAI Slashes Funding Target by $10 Billion Amid Tech Rout (Reuters) – OpenAI cut its fundraising target to roughly $20 billion (down from $30 billion) and its valuation to $1.25 trillion (down from $1.4 trillion), and is rushing to close before Anthropic lists because both are chasing the same institutional investors. It has also opened a tranche to wealthy individuals–a sign that the pool of large investors willing to fund frontier AI at these valuations may be thinner than the headline numbers suggest.

You received this message because you are subscribed to ARPU newsletter - you may unsubscribe anytime at the bottom of this email.

If a friend forwarded you this message, sign up here to get it in your inbox.